Enhancing Essay Grading Efficiency and Consistency through Two-Layer LSTM Models and Attention Mechanisms

Abstract

Pandemic has led people to sought easier solutions for assessing tasks given to students and grading them.  As a potential solution to this problem, Automatic Essay Grading (AEG) has gained considerable attention in recent years. The Kaggle ASAP-AES dataset and the Long Short-Term Memory (LSTM) model with two layers are utilised in this research. The dataset contains essays from a standardised test, providing a more realistic and challenging scenario for AutoGrader. The characteristics from essays are extracted using methods like word tokenization and Feature Vector Creation, pre-processing procedures like stemming and stop-word removal were used. The training data is considered, and an implementation of an average of five folds, consisting of 50 epochs, is made to help the model better focus on pertinent passages in the essay, attention processes was also included. Proposed method provides valuable insights into the potential of AutoGrader to improve the efficiency of and consistency of essay grading, and highlights the effectiveness of the two-layer LSTM model. Experimental findings demonstrate that the two-layer LSTM model surpasses other models in terms of accuracy and efficiency making it a viable strategy for computerized essay scoring systems.